3D Classification
Track 1: 3D classification (No Feat)
In this track, the models are trained only using voxel coordinates on PeRFception-CO3D. In other words, using spherical harmonic coefficients and density values are prohibited. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.
| Model | Acc@1 | Acc@5 | Checkpoint | Code |
|---|---|---|---|---|
| Mink-ResNet101 | 66.21 \(\pm\) 0.93 | 86.50 \(\pm\) 0.09 | link | link |
| Mink-ResNet50 | 65.25 \(\pm\) 0.75 | 85.71 \(\pm\) 0.64 | link | link |
| Mink-ResNet34 | 64.55 \(\pm\) 0.84 | 84.26 \(\pm\) 0.47 | link | link |
| Mink-ResNet18 | 63.85 \(\pm\) 0.33 | 84.47 \(\pm\) 0.54 | link | link |
| Mink-ResNet14 | 59.36 \(\pm\) 0.30 | 81.30 \(\pm\) 0.87 | link | link |
Track 2: 3D classification (SH)
In this track, the models are trained with spherical harmonic coefficients and voxel coordinates on PeRFception-CO3D. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.
| Model | Acc@1 | Acc@5 | Checkpoint | Code |
|---|---|---|---|---|
| Mink-ResNet101 | 78.04 \(\pm\) 0.58 | 89.60 \(\pm\) 0.36 | link | link |
| Mink-ResNet50 | 77.59 \(\pm\) 0.17 | 90.98 \(\pm\) 0.40 | link | link |
| Mink-ResNet34 | 76.51 \(\pm\) 0.61 | 91.79 \(\pm\) 0.49 | link | link |
| Mink-ResNet18 | 75.58 \(\pm\) 0.37 | 92.73 \(\pm\) 0.21 | link | link |
| Mink-ResNet14 | 71.87 \(\pm\) 0.61 | 93.24 \(\pm\) 0.49 | link | link |
Track 3: 3D classification (D)
In this track, the models are trained with density values and voxel coordinates on PeRFception-CO3D. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.
| Model | Acc@1 | Acc@5 | Checkpoint | Code |
|---|---|---|---|---|
| Mink-ResNet101 | 77.27 \(\pm\) 0.61 | 92.68 \(\pm\) 0.34 | link | link |
| Mink-ResNet50 | 76.42 \(\pm\) 0.19 | 92.91 \(\pm\) 0.24 | link | link |
| Mink-ResNet34 | 76.38 \(\pm\) 0.34 | 91.79 \(\pm\) 0.61 | link | link |
| Mink-ResNet18 | 75.18 \(\pm\) 0.70 | 91.10 \(\pm\) 0.24 | link | link |
| Mink-ResNet14 | 72.44 \(\pm\) 0.29 | 90.04 \(\pm\) 0.13 | link | link |
Track 4: 3D classification (SH + D)
In this track, spherical harmonic coefficients, density, and voxel coordinates are available. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.
| Model | Acc@1 | Acc@5 | Checkpoint | Code |
|---|---|---|---|---|
| Mink-ResNet50 | 77.53 \(\pm\) 0.27 | 92.93 \(\pm\) 0.54 | link | link |
| Mink-ResNet101 | 77.19 \(\pm\) 0.89 | 93.09 \(\pm\) 0.21 | link | link |
| Mink-ResNet34 | 76.50 \(\pm\) 0.03 | 91.98 \(\pm\) 0.08 | link | link |
| Mink-ResNet18 | 75.72 \(\pm\) 0.25 | 91.54 \(\pm\) 0.21 | link | link |
| Mink-ResNet14 | 72.92 \(\pm\) 0.42 | 90.83 \(\pm\) 0.03 | link | link |